News AMD Posts Strong Results on Robust EPYC Sales As Consumer CPU Sales Disappoint

[bit_user]

arm is for a completely different workload than what people would buy x86 for,

ARM can basically do it all, now that ARMv9-A added memory encryption and SVE2. More than half of Amazon's AWS fleet is now ARM. ARM was already used for HPC by Fujitsu, about 5 years ago, and is currently being used by the European Processor Initiative:

General Purpose Processor - European Processor Initiative

www.european-processor-initiative.eu

Nvidia is using it as an AI/HPC host processor, in their Grace CPU.

And ARM Neoverse also has a low-power product line, using more efficiency-oriented cores, aimed at the CSP market (e.g. 5G basestations).

That's just servers. It's also being used in self-driving SoCs, robotics, IoT, etc. Basically, they're trying to leave no market untapped.

amd has almost zero support so if you buy from them you are on your own to figure any issues out,

There seems to have been some truth to this, but they've been aggressively hiring to scale up their software & support teams for years, now.

intel has support out the wazoo for both software and hardware.

Yes, the few dealings I've had with Intel's partner support folks have been impressive. Competent people who are responsive and know their stuff. No complaints, there.

ARM can't do everything, nvidia AI can't do everything, amd can't do everything and intel also can't do everything.

Right. Nobody is saying Intel will soon have 0% market share. The problem is they've previously captured so much, and now they're getting squeezed from multiple sides. I think it will definitely decline more, before they're able to mount an effective counteroffensive.

 

[TerryLaze]

Right. Nobody is saying Intel will soon have 0% market share. The problem is they've previously captured so much, and now they're getting squeezed from multiple sides. I think it will definitely decline more, before they're able to mount an effective counteroffensive.

They had only captured a lot of the x86 market,
because they were one of the only two real options, and they only get squeezed by the little bit of squeezing amd can do.
They never had a lot (if they ever had any) of the AI/GPU market or the very high efficiency arm market, they are only now (last few years) investing toward those markets, at least the ai/gpu markets.

 

[bit_user]

They had only captured a lot of the x86 market,

Not only. 10-15 years ago, SPARC, POWER, PowerPC, and MIPS had a share of some markets Intel now dominates. Communications is one example. HPC would be another. Probably SAN is another.

The other thing that happened is that cloud computing, which Intel dominated, has eaten up some of the other markets where non-x86 CPUs played.

 

[digitalgriffin]

Not only. 10-15 years ago, SPARC, POWER, PowerPC, and MIPS had a share of some markets Intel now dominates. Communications is one example. HPC would be another. Probably SAN is another.

The other thing that happened is that cloud computing, which Intel dominated, has eaten up some of the other markets where non-x86 CPUs played.

Yep. Couldn't agree more. Another wildcard if .NET Core v6 and v7. It works on win boxes and multiple flavors Linux. 1 code base multiple compile targets. And that means ARM as well. Even Amazon AWS uses .NET Core. That's how powerful a language it is.

Microsoft saw the writing on the wall in 2007. Cloud would replace the desktop OS boxes as we knew them. And if you look at this video it shows that MS's stranglehold (90%+ of desktop) is no more because of the cloud, where any platform can run it. That includes ARM. This was why MS invested in cloud so much (and a very large portion of their income comes from). MS had adapted to the threat of ARM. Intel has not.

View: https://youtu.be/o14-gCNRwR8

 

[bit_user]

MS had adapted to the threat of ARM. Intel has not.

ARM was never really a threat to MS. Back when Windows NT launched, they even had a port of it to MIPS and DEC Alpha. Maybe PowerPC, too? Then, when Itanium came along, they supported it for a while. MS was never really that wed to x86.

The way I see it, MS viewed Linux somewhat similarly to how Intel views ARM. Eventually MS decided it didn't need to control the OS, in order to control the rest of the stack. Intel doesn't have that same flexibility.

Intel could buy an ARM architectural license and design its own ARM cores, but the downside is that instead of worrying about 1 main competitor on the same ISA, there'd be a huge opening and leveling of the competitive landscape. So, it behooves Intel to stick with x86 for as long as it's still competitive, and only jump to ARM or RISC-V when the market passes a tipping point.

 

[-Fran-]

Yep. Couldn't agree more. Another wildcard if .NET Core v6 and v7. It works on win boxes and multiple flavors Linux. 1 code base multiple compile targets. And that means ARM as well. Even Amazon AWS uses .NET Core. That's how powerful a language it is.

Microsoft saw the writing on the wall in 2007. Cloud would replace the desktop OS boxes as we knew them. And if you look at this video it shows that MS's stranglehold (90%+ of desktop) is no more because of the cloud, where any platform can run it. That includes ARM. This was why MS invested in cloud so much (and a very large portion of their income comes from). MS had adapted to the threat of ARM. Intel has not.

View: https://youtu.be/o14-gCNRwR8

Well, there's a reason why Java became ubiquitous over the years: it's complete independence over the hardware thanks to the virtual machine promise. Microsoft came very late to the "abstraction" party, but they're (unfortunately) winning ground there.

As for the server discussion around profits: unless we're talking about time/mission critical or super specialised/custom, I don't see how Intel can maintain their ground with SPR. I just don't.

Regards.

 

[bit_user]

Microsoft came very late to the "abstraction" party, but they're (unfortunately) winning ground there.

Huh? They tried MS Java, way back in the 90's, but C# and .Net date back to about 2001 or so. At the time, I remember there was lots of confusion in the industry about exactly what .Net was and why Microsoft was doing it.

 

[-Fran-]

Huh? They tried MS Java, way back in the 90's, but C# and .Net date back to about 2001 or so.

Yes, a lot of years after Java got a stronghold of the market. MS couldn't make Sun bend to their will, so...

Ironically, Google could and then Oracle came and well... Hm... Funny.

Regards.

 

[TerryLaze]

Not only. 10-15 years ago, SPARC, POWER, PowerPC, and MIPS had a share of some markets Intel now dominates.

Yeah, so you do agree that intel had strong competition before now.
IBM alone had about 6bil annual revenue in hardware until the end of 2019 which is just 3 years ago and not 10-15.
That's about 25% of the ~21bil intel did in 2019 in data center group.

 

[digitalgriffin]

Well, there's a reason why Java became ubiquitous over the years: it's complete independence over the hardware thanks to the virtual machine promise. Microsoft came very late to the "abstraction" party, but they're (unfortunately) winning ground there.

As for the server discussion around profits: unless we're talking about time/mission critical or super specialised/custom, I don't see how Intel can maintain their ground with SPR. I just don't.

Regards.

Java became the defective because Linux was almost free for universities for a long time. They needed a language that could run on a Linux VM. Java was a natural fit. So was python. Python is most popular, followed by Java, and .net. C++ is a distant 4th because it is platform dependent even with ANSI standards.

Now sun microsystems is charging enterprise users for Java that's starting to change.

 

[-Fran-]

Java became the defective because Linux was almost free for universities for a long time. They needed a language that could run on a Linux VM. Java was a natural fit. So was python. Python is most popular, followed by Java, and .net. C++ is a distant 4th because it is platform dependent even with ANSI standards.

Now sun microsystems is charging enterprise users for Java that's starting to change.

Not quite. OpenJDK and their IBM versions of AIX systems are picking up the slack. For license-driven Corps, you have Zulu, so there's plenty alternatives to Oracle's "official" JDK. That's why Java is not going to die any time soon xD

In other topics:

View: https://www.youtube.com/watch?v=UUvhBE-37SI


Ian's complete breakdown and analysis is masterful.

Regards.

 

[bit_user]

Python is most popular, followed by Java, and .net. C++ is a distant 4th

According to whom? This shows C++ barely edging out Java for 3rd place, with more than double the usage of 5th place C#.

TIOBE Index - TIOBE

www.tiobe.com

because it is platform dependent even with ANSI standards.

C++ is quite portable, if you use libraries like boost or Qt. However, recent versions of the language standard have subsumed much of the core parts of boost, and now even include a platform-independent GUI toolkit.

C++ has undergone something of a renaissance since C++11.

C++ Renaissance

Dynamic language developers who are concerned about performance end up writing pieces of their applications in C++. So if you're going to write C++ anyway, why not write your entire application in C++? Library writers develop in C++ so that their users won't have to. That makes a lot of sense...

www.johndcook.com

If you last used C++ before 2011, you probably wouldn't even recognize it now.

FWIW, here are two more datapoints which rank it #5 and #9 - the PYPL and StackOverflow survey, respectively.

https://www.stackscale.com/blog/most-popular-programming-languages/

 

[-Fran-]

Holy cow, this was so well timed I'm baffled:

View: https://www.youtube.com/watch?v=Nh5i5obZ8OY


Absolutely worth a watch.

Regards.

 

[bit_user]

Holy cow, this was so well timed I'm baffled:

Absolutely worth a watch.

Does it have anything to do with the subjects being discussed in this thread?

 

[-Fran-]

Does it have anything to do with the subjects being discussed in this thread?

Indirectly, yes.

Simply put, if Microsoft and Google (among others, like IBM and Amazon) enter a proper "AI wars", it'll mean more processing power in the backend and it'll mean shifts in how the data is processed and stored in order to optimize for such workloads.

This will, for sure, impact AMD's and Intel's bottom line. Google has already invested in its own hardware designs, Amazon as well and IBM for sure. I'm not sure about MS, but I'd imagine they have?

In all fairness, that video fits several threads that touch on server and profits. Indirectly though.

EDIT: Directly related to AMD's financial reports!

View: https://www.youtube.com/watch?v=3-nAnb_kIko


xD

Regards.

 

[digitalgriffin]

According to whom? This shows C++ barely edging out Java for 3rd place, with more than double the usage of 5th place C#.

TIOBE Index - TIOBE

www.tiobe.com

C++ is quite portable, if you use libraries like boost or Qt. However, recent versions of the language standard have subsumed much of the core parts of boost, and now even include a platform-independent GUI toolkit.

C++ has undergone something of a renaissance since C++11.

C++ Renaissance

Dynamic language developers who are concerned about performance end up writing pieces of their applications in C++. So if you're going to write C++ anyway, why not write your entire application in C++? Library writers develop in C++ so that their users won't have to. That makes a lot of sense...

www.johndcook.com

If you last used C++ before 2011, you probably wouldn't even recognize it now.

FWIW, here are two more datapoints which rank it #5 and #9 - the PYPL and StackOverflow survey, respectively.

https://www.stackscale.com/blog/most-popular-programming-languages/

Turns out we were both have different data. JavaScript is #2...that made me shudder. I do recommend typescript as it's replacement.

View: https://youtu.be/qQXXI5QFUfw

 

[bit_user]

and IBM for sure.

Really? IBM added deep learning hardware to their latest mainframe CPUs, but those are far from cost-effective. I think that was done only to enable existing mainframe customers to do some deep learning without needing to downgrade their hardware standards. For instance, banks might have a policy or regulatory requirement to use mainframes for their realtime transaction processing, but they also want to do fraud-detection on those transactions. At least, that's the only way I can make any sense of it, because those mainframes make Nvidia's DGX systems look cheap by comparison (and Nvidia is orders of magnitude faster).

Did IBM build any other deep learning hardware?

I'm not sure about MS, but I'd imagine they have?

It's plausible, but let's try and stick to what we actually know.

 

[-Fran-]

Really? IBM added deep learning hardware to their latest mainframe CPUs, but those are far from cost-effective. I think that was done only to enable existing mainframe customers to do some deep learning without needing to downgrade their hardware standards. For instance, banks might have a policy or regulatory requirement to use mainframes for their realtime transaction processing, but they also want to do fraud-detection on those transactions. At least, that's the only way I can make any sense of it, because those mainframes make Nvidia's DGX systems look cheap by comparison (and Nvidia is orders of magnitude faster).

Did IBM build any other deep learning hardware?


It's plausible, but let's try and stick to what we actually know.

I mentioned IBM as a "legacy" thing. Remember "Deep Blue"*?

If I'm not wrong, IBM still offers a medical AI assistant before "AI" was a thing. Whether or not it uses specialised hardware, I don't remember. I do remember their first versions used PowerPC CPUs to power them.

And yeah, I have no idea about MS. I know they have the resources to implement something custom, but I don't know if they have or have plans.

Regards.

 

[bit_user]

I mentioned IBM as a "legacy" thing. Remember "Deep Blue"*?

If I'm not wrong, IBM still offers a medical AI assistant before "AI" was a thing. Whether or not it uses specialised hardware, I don't remember. I do remember their first versions used PowerPC CPUs to power them.

And yeah, I have no idea about MS. I know they have the resources to implement something custom, but I don't know if they have or have plans.

I thought we were talking specifically about current AI accelerator chips or blocks.

And, on that note, it looks like Microsoft's bespoke solutions are FPGA-based:

AI at Scale - Microsoft Research

AI at Scale is an applied research initiative that works to evolve Microsoft products with the adoption of deep learning for both natural language text and image processing. Our work is actively being integrated into Microsoft products, including Bing, Office, and Xbox.

www.microsoft.com

...good for some things, but I think not competitive with the best-in-class AI ASICs at the sort of tasks for which they're optimized. Maybe for related reasons, Microsoft offers GPU-based Azure instances.

 

[msroadkill612]

Except the egg price spike is due to avian influenza, and therefore shouldn't be taken as an indicator of inflation. Once that passes and hen stocks have a chance to recover, you'll see egg prices return to long-term trends.


In terms of gross numbers, the layoffs announced so far haven't really made a dent in the overall unemployment rate. However, they are focused on a mid/high-income, tech-oriented segment of the workforce, which tends to do a disproportionate amount of the spending on tech products. Furthermore, the rest of that workforce that's still employed will be more wary of making big, unnecessary purchases on PCs and accessories, until the overall health of this sector improves. Too often, one round of layoffs is followed by another.

More importantly, corporate spending on IT products will also be down.

FYI, automotive is soaking up the tech layoffs. They dont want another apple doing to autos what they did to phones.

They want to own their platform.

 

[Deleted member 14196]

All the new Teslas use AMD Apus now. There’s a huge source of revenue right there in automotive especially as Tesla scales up to their goal which is huge

 

[TerryLaze]

All the new Teslas use AMD Apus now. There’s a huge source of revenue right there in automotive especially as Tesla scales up to their goal which is huge

I wouldn't bet my horse on it being huge...big maybe.
The margins on these kinds of deals are usually small and the production capability for these APUs is small as well.
These APUs weren't enough for the consoles alone for the last 2 years.

 

[bit_user]

All the new Teslas use AMD Apus now.

Only for the entertainment system. Tesla makes its own SoC for self-driving and presumably the new humanoid robots they announced.